Open skills, protocols, plugins, and integration recipes for the agentbox ecosystem — the parts that only work when open.
Status: early stage, slow-updating.
agentbox is the master brand, positioned as infrastructure for agents. The current product is agentbox.id, a minimal mailbox service for AI agents — a Guardian (human owner) verifies once with their existing email and gets up to 20 agent mailboxes with standard IMAP/SMTP credentials, consumable by Hermes Agent, openclaw, GenericAgent, and any other runtime that speaks email.
A sibling product, soul.agentbox.id, is in early validation; it does not depend on the mailbox service and is not gated by it.
This repo holds the parts of the agentbox ecosystem that are only useful when open:
- Skills — the AI-native artifact. An agent reads a skill's full text before executing against it; closed skills are non-starters in any autonomous-agent workflow. Following the
SKILL.md+references/+scripts/+assets/convention shared with openai/skills, anthropics/skills, and the broader Hermes ecosystem. - Protocols — RFC-style specifications that closed-source product implementations answer to.
- Plugins — runtime-side glue that agent runtimes install to integrate with agentbox services.
- Integration recipes — how to adapt Hermes Agent / openclaw / GenericAgent / etc. so they can use agentbox.
- Background essays — the thinking behind the work.
The principle is straightforward: a closed skill or plugin is an agent being asked to execute opaque instructions on its owner's behalf, which most autonomous-agent operators won't accept. The artifacts that make agentbox feel like an AI-native project — and the artifacts users need to inspect before trusting — are the same artifacts. So they live here, in the open.
- Not the source code of agentbox.id. The product itself is closed.
- Not a framework. We publish contracts, not runtimes.
- Not a finished standard. Everything here is a working draft, shaped by use rather than committee.
docs/— documentation hub. Background essays, and (coming) product rules, integration guides, setup walkthroughs. Will be deployed atdocs.agentbox.id.protocols/— RFC-like protocol specifications. Currently: Agent Attention Runtime.skills/— capability units consumable by agents. Following theSKILL.mdconvention.plugins/— agentbox-provided plug-ins that agent runtimes install. Placeholder; the concept will sharpen as we build the first one.examples/— practical usage examples.
Three principles inform the work:
- Token economics — every agent action has a cost; someone must pay.
- Conversation rhythm — silence, delay, and cadence are signals, not gaps.
- Guardianship — agents are not yet full legal actors; responsibility must trace to someone.
Long form, for readers who want the full case: docs/background/.
This is a thinking-and-building repository, not a product roadmap. Updates here lag the product on purpose: what we publish is validated abstractions in progress, not a feed of every change shipped to agentbox.id.
Issues and discussions welcome; expect replies in days, not hours.
If you want to use a skill, protocol, or plugin from here in something serious, pin a version. Things will change.
Licensed under the Apache License 2.0. See NOTICE for attribution details.
This repository is maintained as part of the agentbox.id ecosystem. Some content here was shaped through dialogue with AI systems; the arguments and judgments are the maintainer's own.